This is the step people skip, and it is the one that costs the most to skip.
Nearly every preprocessing pipeline worth using — fMRIPrep, QSIPrep, MRIQC — takes BIDS as its input format. If your data is in BIDS, those tools run with almost no configuration. If it is not, you will either write conversion glue for each one or give up on them.
BIDS¶
The Brain Imaging Data Structure — what the specification is, why it makes data FAIR, and the surrounding tool ecosystem
BIDS is a filesystem convention, not a file format: a directory layout, consistent naming, and JSON sidecars carrying acquisition metadata. Extensions cover EEG, MEG, and PET, so the same conventions apply beyond MRI.
Practical advice from people who have done it: convert one subject by hand before you automate anything. The validator is strict and its errors assume you already know the specification. An hour spent understanding the layout on a single subject saves days of fighting a conversion script.
Managing a whole dataset¶
One subject is a conversion problem; a hundred is a bookkeeping problem — which subjects have been processed, with which pipeline version, and which failed.
Nipoppy — a framework for organizing and tracking pipeline runs across a full dataset, rather than invoking tools per subject
Keeping track of the rest¶
Project management — organizing the project around the data
Zettelkasten — a note-taking method for keeping research notes navigable over a multi-year project
Two related pages: getting data into BIDS for the conversion itself, and DataLad for versioning the dataset while you do it.
Next: Preprocess.